Skip to content

Commit bdcb512

Browse files
Mel-ChenMel-Chen
committed
UF2 test
1 parent 4ae69dc commit bdcb512

File tree

1 file changed

+212
-9
lines changed

1 file changed

+212
-9
lines changed

‎llvm/test/Transforms/LoopVectorize/RISCV/riscv-vector-reverse-output.ll

Lines changed: 212 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -7,13 +7,17 @@
77
;; a[i] = b[i] + 1.0;
88

99
; RUN: opt -passes=loop-vectorize -mtriple=riscv64 -mattr=+v \
10-
; RUN: -scalable-vectorization=on -riscv-v-vector-bits-min=128 -S < %s \
10+
; RUN: -riscv-v-vector-bits-min=128 -S < %s \
1111
; RUN: | FileCheck --check-prefix=RV64 %s
1212

1313
; RUN: opt -passes=loop-vectorize -mtriple=riscv32 -mattr=+v \
14-
; RUN: -scalable-vectorization=on -riscv-v-vector-bits-min=128 -S < %s \
14+
; RUN: -riscv-v-vector-bits-min=128 -S < %s \
1515
; RUN: | FileCheck --check-prefix=RV32 %s
1616

17+
; RUN: opt -passes=loop-vectorize -mtriple=riscv64 -mattr=+v \
18+
; RUN: -riscv-v-vector-bits-min=128 -force-vector-interleave=2 -S < %s \
19+
; RUN: | FileCheck --check-prefix=RV64-UF2 %s
20+
1721
define void @vector_reverse_i64(ptr noalias %A, ptr noalias %B, i32 %n) {
1822
; RV64-LABEL: define void @vector_reverse_i64(
1923
; RV64-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i32 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
@@ -174,6 +178,105 @@ define void @vector_reverse_i64(ptr noalias %A, ptr noalias %B, i32 %n) {
174178
; RV32-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
175179
; RV32-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP4:![0-9]+]]
176180
;
181+
; RV64-UF2-LABEL: define void @vector_reverse_i64(
182+
; RV64-UF2-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i32 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
183+
; RV64-UF2-NEXT: [[ENTRY:.*:]]
184+
; RV64-UF2-NEXT: [[CMP7:%.*]] = icmp sgt i32 [[N]], 0
185+
; RV64-UF2-NEXT: br i1 [[CMP7]], label %[[FOR_BODY_PREHEADER:.*]], label %[[FOR_COND_CLEANUP:.*]]
186+
; RV64-UF2: [[FOR_BODY_PREHEADER]]:
187+
; RV64-UF2-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64
188+
; RV64-UF2-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
189+
; RV64-UF2-NEXT: [[TMP2:%.*]] = mul i64 [[TMP1]], 8
190+
; RV64-UF2-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], [[TMP2]]
191+
; RV64-UF2-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
192+
; RV64-UF2: [[VECTOR_SCEVCHECK]]:
193+
; RV64-UF2-NEXT: [[TMP3:%.*]] = add nsw i64 [[TMP0]], -1
194+
; RV64-UF2-NEXT: [[TMP4:%.*]] = add i32 [[N]], -1
195+
; RV64-UF2-NEXT: [[TMP5:%.*]] = trunc i64 [[TMP3]] to i32
196+
; RV64-UF2-NEXT: [[MUL:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 1, i32 [[TMP5]])
197+
; RV64-UF2-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL]], 0
198+
; RV64-UF2-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL]], 1
199+
; RV64-UF2-NEXT: [[TMP6:%.*]] = sub i32 [[TMP4]], [[MUL_RESULT]]
200+
; RV64-UF2-NEXT: [[TMP7:%.*]] = icmp ugt i32 [[TMP6]], [[TMP4]]
201+
; RV64-UF2-NEXT: [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW]]
202+
; RV64-UF2-NEXT: [[TMP9:%.*]] = icmp ugt i64 [[TMP3]], 4294967295
203+
; RV64-UF2-NEXT: [[TMP10:%.*]] = or i1 [[TMP8]], [[TMP9]]
204+
; RV64-UF2-NEXT: br i1 [[TMP10]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
205+
; RV64-UF2: [[VECTOR_PH]]:
206+
; RV64-UF2-NEXT: [[TMP11:%.*]] = call i64 @llvm.vscale.i64()
207+
; RV64-UF2-NEXT: [[TMP12:%.*]] = mul i64 [[TMP11]], 8
208+
; RV64-UF2-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], [[TMP12]]
209+
; RV64-UF2-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
210+
; RV64-UF2-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64()
211+
; RV64-UF2-NEXT: [[TMP14:%.*]] = mul i64 [[TMP13]], 4
212+
; RV64-UF2-NEXT: [[TMP15:%.*]] = mul i64 [[TMP14]], 2
213+
; RV64-UF2-NEXT: [[TMP16:%.*]] = sub i64 [[TMP0]], [[N_VEC]]
214+
; RV64-UF2-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i32
215+
; RV64-UF2-NEXT: [[TMP17:%.*]] = sub i32 [[N]], [[DOTCAST]]
216+
; RV64-UF2-NEXT: br label %[[VECTOR_BODY:.*]]
217+
; RV64-UF2: [[VECTOR_BODY]]:
218+
; RV64-UF2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
219+
; RV64-UF2-NEXT: [[DOTCAST1:%.*]] = trunc i64 [[INDEX]] to i32
220+
; RV64-UF2-NEXT: [[OFFSET_IDX:%.*]] = sub i32 [[N]], [[DOTCAST1]]
221+
; RV64-UF2-NEXT: [[TMP18:%.*]] = add i32 [[OFFSET_IDX]], 0
222+
; RV64-UF2-NEXT: [[TMP19:%.*]] = add nsw i32 [[TMP18]], -1
223+
; RV64-UF2-NEXT: [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
224+
; RV64-UF2-NEXT: [[TMP21:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP20]]
225+
; RV64-UF2-NEXT: [[TMP22:%.*]] = mul i64 0, [[TMP14]]
226+
; RV64-UF2-NEXT: [[TMP23:%.*]] = sub i64 1, [[TMP14]]
227+
; RV64-UF2-NEXT: [[TMP24:%.*]] = getelementptr inbounds i32, ptr [[TMP21]], i64 [[TMP22]]
228+
; RV64-UF2-NEXT: [[TMP25:%.*]] = getelementptr inbounds i32, ptr [[TMP24]], i64 [[TMP23]]
229+
; RV64-UF2-NEXT: [[TMP26:%.*]] = mul i64 -1, [[TMP14]]
230+
; RV64-UF2-NEXT: [[TMP27:%.*]] = sub i64 1, [[TMP14]]
231+
; RV64-UF2-NEXT: [[TMP28:%.*]] = getelementptr inbounds i32, ptr [[TMP21]], i64 [[TMP26]]
232+
; RV64-UF2-NEXT: [[TMP29:%.*]] = getelementptr inbounds i32, ptr [[TMP28]], i64 [[TMP27]]
233+
; RV64-UF2-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP25]], align 4
234+
; RV64-UF2-NEXT: [[REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[WIDE_LOAD]])
235+
; RV64-UF2-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x i32>, ptr [[TMP29]], align 4
236+
; RV64-UF2-NEXT: [[REVERSE3:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[WIDE_LOAD2]])
237+
; RV64-UF2-NEXT: [[TMP30:%.*]] = add <vscale x 4 x i32> [[REVERSE]], splat (i32 1)
238+
; RV64-UF2-NEXT: [[TMP31:%.*]] = add <vscale x 4 x i32> [[REVERSE3]], splat (i32 1)
239+
; RV64-UF2-NEXT: [[TMP32:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP20]]
240+
; RV64-UF2-NEXT: [[TMP33:%.*]] = mul i64 0, [[TMP14]]
241+
; RV64-UF2-NEXT: [[TMP34:%.*]] = sub i64 1, [[TMP14]]
242+
; RV64-UF2-NEXT: [[TMP35:%.*]] = getelementptr inbounds i32, ptr [[TMP32]], i64 [[TMP33]]
243+
; RV64-UF2-NEXT: [[TMP36:%.*]] = getelementptr inbounds i32, ptr [[TMP35]], i64 [[TMP34]]
244+
; RV64-UF2-NEXT: [[TMP37:%.*]] = mul i64 -1, [[TMP14]]
245+
; RV64-UF2-NEXT: [[TMP38:%.*]] = sub i64 1, [[TMP14]]
246+
; RV64-UF2-NEXT: [[TMP39:%.*]] = getelementptr inbounds i32, ptr [[TMP32]], i64 [[TMP37]]
247+
; RV64-UF2-NEXT: [[TMP40:%.*]] = getelementptr inbounds i32, ptr [[TMP39]], i64 [[TMP38]]
248+
; RV64-UF2-NEXT: [[REVERSE4:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[TMP30]])
249+
; RV64-UF2-NEXT: store <vscale x 4 x i32> [[REVERSE4]], ptr [[TMP36]], align 4
250+
; RV64-UF2-NEXT: [[REVERSE5:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[TMP31]])
251+
; RV64-UF2-NEXT: store <vscale x 4 x i32> [[REVERSE5]], ptr [[TMP40]], align 4
252+
; RV64-UF2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP15]]
253+
; RV64-UF2-NEXT: [[TMP41:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
254+
; RV64-UF2-NEXT: br i1 [[TMP41]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
255+
; RV64-UF2: [[MIDDLE_BLOCK]]:
256+
; RV64-UF2-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
257+
; RV64-UF2-NEXT: br i1 [[CMP_N]], label %[[FOR_COND_CLEANUP_LOOPEXIT:.*]], label %[[SCALAR_PH]]
258+
; RV64-UF2: [[SCALAR_PH]]:
259+
; RV64-UF2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP16]], %[[MIDDLE_BLOCK]] ], [ [[TMP0]], %[[FOR_BODY_PREHEADER]] ], [ [[TMP0]], %[[VECTOR_SCEVCHECK]] ]
260+
; RV64-UF2-NEXT: [[BC_RESUME_VAL6:%.*]] = phi i32 [ [[TMP17]], %[[MIDDLE_BLOCK]] ], [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_SCEVCHECK]] ]
261+
; RV64-UF2-NEXT: br label %[[FOR_BODY:.*]]
262+
; RV64-UF2: [[FOR_COND_CLEANUP_LOOPEXIT]]:
263+
; RV64-UF2-NEXT: br label %[[FOR_COND_CLEANUP]]
264+
; RV64-UF2: [[FOR_COND_CLEANUP]]:
265+
; RV64-UF2-NEXT: ret void
266+
; RV64-UF2: [[FOR_BODY]]:
267+
; RV64-UF2-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ]
268+
; RV64-UF2-NEXT: [[I_0_IN8:%.*]] = phi i32 [ [[BC_RESUME_VAL6]], %[[SCALAR_PH]] ], [ [[I_0:%.*]], %[[FOR_BODY]] ]
269+
; RV64-UF2-NEXT: [[I_0]] = add nsw i32 [[I_0_IN8]], -1
270+
; RV64-UF2-NEXT: [[IDXPROM:%.*]] = zext i32 [[I_0]] to i64
271+
; RV64-UF2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IDXPROM]]
272+
; RV64-UF2-NEXT: [[TMP42:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
273+
; RV64-UF2-NEXT: [[ADD9:%.*]] = add i32 [[TMP42]], 1
274+
; RV64-UF2-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IDXPROM]]
275+
; RV64-UF2-NEXT: store i32 [[ADD9]], ptr [[ARRAYIDX3]], align 4
276+
; RV64-UF2-NEXT: [[CMP:%.*]] = icmp ugt i64 [[INDVARS_IV]], 1
277+
; RV64-UF2-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
278+
; RV64-UF2-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP4:![0-9]+]]
279+
;
177280
entry:
178281
%cmp7 = icmp sgt i32 %n, 0
179282
br i1 %cmp7, label %for.body.preheader, label %for.cond.cleanup
@@ -360,6 +463,105 @@ define void @vector_reverse_f32(ptr noalias %A, ptr noalias %B, i32 %n) {
360463
; RV32-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
361464
; RV32-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP6:![0-9]+]]
362465
;
466+
; RV64-UF2-LABEL: define void @vector_reverse_f32(
467+
; RV64-UF2-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i32 [[N:%.*]]) #[[ATTR0]] {
468+
; RV64-UF2-NEXT: [[ENTRY:.*:]]
469+
; RV64-UF2-NEXT: [[CMP7:%.*]] = icmp sgt i32 [[N]], 0
470+
; RV64-UF2-NEXT: br i1 [[CMP7]], label %[[FOR_BODY_PREHEADER:.*]], label %[[FOR_COND_CLEANUP:.*]]
471+
; RV64-UF2: [[FOR_BODY_PREHEADER]]:
472+
; RV64-UF2-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64
473+
; RV64-UF2-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64()
474+
; RV64-UF2-NEXT: [[TMP2:%.*]] = mul i64 [[TMP1]], 8
475+
; RV64-UF2-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], [[TMP2]]
476+
; RV64-UF2-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
477+
; RV64-UF2: [[VECTOR_SCEVCHECK]]:
478+
; RV64-UF2-NEXT: [[TMP3:%.*]] = add nsw i64 [[TMP0]], -1
479+
; RV64-UF2-NEXT: [[TMP4:%.*]] = add i32 [[N]], -1
480+
; RV64-UF2-NEXT: [[TMP5:%.*]] = trunc i64 [[TMP3]] to i32
481+
; RV64-UF2-NEXT: [[MUL:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 1, i32 [[TMP5]])
482+
; RV64-UF2-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL]], 0
483+
; RV64-UF2-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL]], 1
484+
; RV64-UF2-NEXT: [[TMP6:%.*]] = sub i32 [[TMP4]], [[MUL_RESULT]]
485+
; RV64-UF2-NEXT: [[TMP7:%.*]] = icmp ugt i32 [[TMP6]], [[TMP4]]
486+
; RV64-UF2-NEXT: [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW]]
487+
; RV64-UF2-NEXT: [[TMP9:%.*]] = icmp ugt i64 [[TMP3]], 4294967295
488+
; RV64-UF2-NEXT: [[TMP10:%.*]] = or i1 [[TMP8]], [[TMP9]]
489+
; RV64-UF2-NEXT: br i1 [[TMP10]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
490+
; RV64-UF2: [[VECTOR_PH]]:
491+
; RV64-UF2-NEXT: [[TMP11:%.*]] = call i64 @llvm.vscale.i64()
492+
; RV64-UF2-NEXT: [[TMP12:%.*]] = mul i64 [[TMP11]], 8
493+
; RV64-UF2-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], [[TMP12]]
494+
; RV64-UF2-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
495+
; RV64-UF2-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64()
496+
; RV64-UF2-NEXT: [[TMP14:%.*]] = mul i64 [[TMP13]], 4
497+
; RV64-UF2-NEXT: [[TMP15:%.*]] = mul i64 [[TMP14]], 2
498+
; RV64-UF2-NEXT: [[TMP16:%.*]] = sub i64 [[TMP0]], [[N_VEC]]
499+
; RV64-UF2-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i32
500+
; RV64-UF2-NEXT: [[TMP17:%.*]] = sub i32 [[N]], [[DOTCAST]]
501+
; RV64-UF2-NEXT: br label %[[VECTOR_BODY:.*]]
502+
; RV64-UF2: [[VECTOR_BODY]]:
503+
; RV64-UF2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
504+
; RV64-UF2-NEXT: [[DOTCAST1:%.*]] = trunc i64 [[INDEX]] to i32
505+
; RV64-UF2-NEXT: [[OFFSET_IDX:%.*]] = sub i32 [[N]], [[DOTCAST1]]
506+
; RV64-UF2-NEXT: [[TMP18:%.*]] = add i32 [[OFFSET_IDX]], 0
507+
; RV64-UF2-NEXT: [[TMP19:%.*]] = add nsw i32 [[TMP18]], -1
508+
; RV64-UF2-NEXT: [[TMP20:%.*]] = zext i32 [[TMP19]] to i64
509+
; RV64-UF2-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP20]]
510+
; RV64-UF2-NEXT: [[TMP22:%.*]] = mul i64 0, [[TMP14]]
511+
; RV64-UF2-NEXT: [[TMP23:%.*]] = sub i64 1, [[TMP14]]
512+
; RV64-UF2-NEXT: [[TMP24:%.*]] = getelementptr inbounds float, ptr [[TMP21]], i64 [[TMP22]]
513+
; RV64-UF2-NEXT: [[TMP25:%.*]] = getelementptr inbounds float, ptr [[TMP24]], i64 [[TMP23]]
514+
; RV64-UF2-NEXT: [[TMP26:%.*]] = mul i64 -1, [[TMP14]]
515+
; RV64-UF2-NEXT: [[TMP27:%.*]] = sub i64 1, [[TMP14]]
516+
; RV64-UF2-NEXT: [[TMP28:%.*]] = getelementptr inbounds float, ptr [[TMP21]], i64 [[TMP26]]
517+
; RV64-UF2-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, ptr [[TMP28]], i64 [[TMP27]]
518+
; RV64-UF2-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x float>, ptr [[TMP25]], align 4
519+
; RV64-UF2-NEXT: [[REVERSE:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD]])
520+
; RV64-UF2-NEXT: [[WIDE_LOAD2:%.*]] = load <vscale x 4 x float>, ptr [[TMP29]], align 4
521+
; RV64-UF2-NEXT: [[REVERSE3:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD2]])
522+
; RV64-UF2-NEXT: [[TMP30:%.*]] = fadd <vscale x 4 x float> [[REVERSE]], splat (float 1.000000e+00)
523+
; RV64-UF2-NEXT: [[TMP31:%.*]] = fadd <vscale x 4 x float> [[REVERSE3]], splat (float 1.000000e+00)
524+
; RV64-UF2-NEXT: [[TMP32:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP20]]
525+
; RV64-UF2-NEXT: [[TMP33:%.*]] = mul i64 0, [[TMP14]]
526+
; RV64-UF2-NEXT: [[TMP34:%.*]] = sub i64 1, [[TMP14]]
527+
; RV64-UF2-NEXT: [[TMP35:%.*]] = getelementptr inbounds float, ptr [[TMP32]], i64 [[TMP33]]
528+
; RV64-UF2-NEXT: [[TMP36:%.*]] = getelementptr inbounds float, ptr [[TMP35]], i64 [[TMP34]]
529+
; RV64-UF2-NEXT: [[TMP37:%.*]] = mul i64 -1, [[TMP14]]
530+
; RV64-UF2-NEXT: [[TMP38:%.*]] = sub i64 1, [[TMP14]]
531+
; RV64-UF2-NEXT: [[TMP39:%.*]] = getelementptr inbounds float, ptr [[TMP32]], i64 [[TMP37]]
532+
; RV64-UF2-NEXT: [[TMP40:%.*]] = getelementptr inbounds float, ptr [[TMP39]], i64 [[TMP38]]
533+
; RV64-UF2-NEXT: [[REVERSE4:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[TMP30]])
534+
; RV64-UF2-NEXT: store <vscale x 4 x float> [[REVERSE4]], ptr [[TMP36]], align 4
535+
; RV64-UF2-NEXT: [[REVERSE5:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[TMP31]])
536+
; RV64-UF2-NEXT: store <vscale x 4 x float> [[REVERSE5]], ptr [[TMP40]], align 4
537+
; RV64-UF2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP15]]
538+
; RV64-UF2-NEXT: [[TMP41:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
539+
; RV64-UF2-NEXT: br i1 [[TMP41]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
540+
; RV64-UF2: [[MIDDLE_BLOCK]]:
541+
; RV64-UF2-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
542+
; RV64-UF2-NEXT: br i1 [[CMP_N]], label %[[FOR_COND_CLEANUP_LOOPEXIT:.*]], label %[[SCALAR_PH]]
543+
; RV64-UF2: [[SCALAR_PH]]:
544+
; RV64-UF2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP16]], %[[MIDDLE_BLOCK]] ], [ [[TMP0]], %[[FOR_BODY_PREHEADER]] ], [ [[TMP0]], %[[VECTOR_SCEVCHECK]] ]
545+
; RV64-UF2-NEXT: [[BC_RESUME_VAL6:%.*]] = phi i32 [ [[TMP17]], %[[MIDDLE_BLOCK]] ], [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_SCEVCHECK]] ]
546+
; RV64-UF2-NEXT: br label %[[FOR_BODY:.*]]
547+
; RV64-UF2: [[FOR_COND_CLEANUP_LOOPEXIT]]:
548+
; RV64-UF2-NEXT: br label %[[FOR_COND_CLEANUP]]
549+
; RV64-UF2: [[FOR_COND_CLEANUP]]:
550+
; RV64-UF2-NEXT: ret void
551+
; RV64-UF2: [[FOR_BODY]]:
552+
; RV64-UF2-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ]
553+
; RV64-UF2-NEXT: [[I_0_IN8:%.*]] = phi i32 [ [[BC_RESUME_VAL6]], %[[SCALAR_PH]] ], [ [[I_0:%.*]], %[[FOR_BODY]] ]
554+
; RV64-UF2-NEXT: [[I_0]] = add nsw i32 [[I_0_IN8]], -1
555+
; RV64-UF2-NEXT: [[IDXPROM:%.*]] = zext i32 [[I_0]] to i64
556+
; RV64-UF2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[IDXPROM]]
557+
; RV64-UF2-NEXT: [[TMP42:%.*]] = load float, ptr [[ARRAYIDX]], align 4
558+
; RV64-UF2-NEXT: [[CONV1:%.*]] = fadd float [[TMP42]], 1.000000e+00
559+
; RV64-UF2-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IDXPROM]]
560+
; RV64-UF2-NEXT: store float [[CONV1]], ptr [[ARRAYIDX3]], align 4
561+
; RV64-UF2-NEXT: [[CMP:%.*]] = icmp ugt i64 [[INDVARS_IV]], 1
562+
; RV64-UF2-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
563+
; RV64-UF2-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP6:![0-9]+]]
564+
;
363565
entry:
364566
%cmp7 = icmp sgt i32 %n, 0
365567
br i1 %cmp7, label %for.body.preheader, label %for.cond.cleanup
@@ -391,13 +593,6 @@ for.body: ; preds = %for.body.preheader,
391593
!2 = !{!"llvm.loop.vectorize.width", i32 4}
392594
!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
393595
!4 = !{!"llvm.loop.vectorize.enable", i1 true}
394-
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]], [[META3:![0-9]+]]}
395-
; CHECK: [[META1]] = !{!"llvm.loop.mustprogress"}
396-
; CHECK: [[META2]] = !{!"llvm.loop.isvectorized", i32 1}
397-
; CHECK: [[META3]] = !{!"llvm.loop.unroll.runtime.disable"}
398-
; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
399-
; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]], [[META2]], [[META3]]}
400-
; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]}
401596
;.
402597
; RV64: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]], [[META3:![0-9]+]]}
403598
; RV64: [[META1]] = !{!"llvm.loop.mustprogress"}
@@ -415,3 +610,11 @@ for.body: ; preds = %for.body.preheader,
415610
; RV32: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]], [[META2]], [[META3]]}
416611
; RV32: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META3]], [[META2]]}
417612
;.
613+
; RV64-UF2: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]], [[META3:![0-9]+]]}
614+
; RV64-UF2: [[META1]] = !{!"llvm.loop.mustprogress"}
615+
; RV64-UF2: [[META2]] = !{!"llvm.loop.isvectorized", i32 1}
616+
; RV64-UF2: [[META3]] = !{!"llvm.loop.unroll.runtime.disable"}
617+
; RV64-UF2: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
618+
; RV64-UF2: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]], [[META2]], [[META3]]}
619+
; RV64-UF2: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]}
620+
;.

0 commit comments

Comments
 (0)